What can we expect from GPT-4?
Having tasted the potential of ChatGPT, users worldwide – want more. The GPT 3.5 currently in use is undoubtedly fascinating; even its shortcomings and limitations were perceived as lovable and entertaining. However, that’s exactly where they should put their focus in the upcoming version, the widely announced GPT-4.
Nevertheless, expecting it to be flawless could only lead to disappointment.
What can we expect from GPT-4?
According to OpenAI CEO Sam Altman:
- The chatbot will have great use in providing medical services, i.e. giving medical advice, but also in education and teaching. We will see an even greater application of the computer as a friend.
- The new model will not be able to generate new knowledge for humanity, lead to new scientific discoveries or cure cancer. Still, it will be able to do what people have already done and learned.
- If we use them correctly, these models can make us twice as productive, introduce new approaches to research that we haven’t used before and make us more time-efficient.
- The dialog form will be perfected and will allow the machine to do things instead of humans, as we have seen in previous versions, for example, in copywriting, but it will now be more precise, with more possibilities for iterations and fine-tuning.
- In the future, there will no longer be so many startups that develop their own AI models, but will make use of the existing ones and adapt them to their needs.
- One of the main goals is for GPT to no longer be limited by the time within which it was trained.
- For the first time, Google will have a serious challenge when it comes to a search product.
What do the experts say?
You can find so many self-proclaimed experts online who gave their opinions on the future of GPT-4, most commonly based on the current hype. Others, however, don’t hide their fears about this technology.
Either way, real experts in this field provide much more realistic predictions.
We discussed this topic with Prof. Dr. Dubravko Ćulibrk, Acting Director of the Institute for Artificial Intelligence of Serbia.
How will GPT-4 differ from previous versions?
– I think that any opinion on this topic is, at its core, mere speculation. I would only rely on what OpenAI founder, Sam Altman, stated, which is that people will be disappointed, because their expectations are too high. If I have to speculate, I would say that I expect growth to slow down in terms of the complexity of the model (the number of artificial neural networks), to somehow incorporate continuous learning (the current model, for example, has very limited knowledge of the world after 2021), to use the innovations introduced for the development of ChatGPT (learning by incentive and through user interaction) to improve the base model itself, and that the GPT-4 will make a leap forward to become a fundamental model that can process not only text, but also image, video, etc.
How do you see the future application of this technology, and in what ways can it improve the lives of users?
– Basically, it’s a technology that allows computers to communicate more efficiently in natural (human) language. In addition to facilitated interaction between people, whether through text or speech, such technologies enable the development of better translation systems from one human language to another, the extraction of information from written texts (such as historical texts or medical reports), but also the development of advanced tools that greatly automate the development of computer search engines. Lastly, it has the potential to democratize software development, and at the same time accelerate application development.
Do you believe that GPT-4 will be used as a search engine (instead of Google, for example), and that users won’t have the need anymore to go to other websites to get their answers?
– In terms of search engines, models from the GPT family allow a better understanding of the query, as well as the exact content on a particular website. Google’s key technology is the page ranking algorithm, which GPT-4 can help with, but, without some major additional development, it can hardly replace it. In other words, it is possible for Google to use GPT to improve this part of its functionality, but I would say that Google has significantly more expertise in artificial intelligence than OpenAI and that, quite certainly, it has its own models of complexity similar to GPT-3, and will continue to develop them. It shouldn’t be forgotten that the first transformation network (BERT), to which the “T” in GPT refers, originates exactly from Google’s labs. What’s a much more likely scenario is that Microsoft will use the license it has for GPT-3 and the licenses Microsoft will certainly have for the new incarnations of this technology to try to conquer most of the Internet browser market.
ChatGPT has become one of the first things that come to mind when we mention artificial intelligence. Have you noticed that the public’s interest in this topic has increased? Is now the right time to destigmatize it, especially among users who were skeptical about this technology?
– There is definitely an increased interest of the general public in this topic. I’m not sure how much ChatGPT has potential in terms of destigmatization, except as a replacement for other buzzwords like Skynet and Terminator. Anyhow, I think it’s a good step forward in this field, but it will also be a long and gradual process.